How do emotions surrounding past instances of intergroup violence influence future conflict? Previous research argues that emotions stemming from past violence—particularly anger—can drive the continuation of conflict (Bar-Tal Reference Bar-Tal2000; Horowitz Reference Horowitz2001). I sought to answer this question using a behavioral economics experiment that examined how experimental primes of anger about past ethnic violence influenced intergroup conflict. The Israeli–Palestinian conflict is one of the longest-running continual conflicts. It is notable for its strong ingroup–outgroup identities, with a history of victimization on both sides that has been used as justification for current violence (Morris Reference Morris2011; Tessler Reference Tessler2009).
I chose to explore the relationship between anger and intergroup violence in Acre, Israel, a mixed Jewish–Palestinian Citizen of Israel (PCI) city of 50,000 people (Israel Central Bureau of Statistics 2011). Approximately two thirds of Acre’s population at the time of my study was Jewish and one third was PCI. The choice of Acre stemmed from both its demographics—as one of the few mixed Jewish–PCI cities in Israel—and as a repeated flashpoint for Jewish–PCI conflict (Torstrick Reference Torstrick2000).
I was particularly interested in studying the effects of the 2008 interethnic violence known as the “Yom Kippur Riots.” The riots began when a young PCI, Tawfik Jamal, drove into a predominantly Jewish neighborhood on Yom Kippur, the holiest day of the year for Jews.Footnote 1 The driver was stopped and assaulted by Jewish youths from a local apartment complex. He then was chased into an apartment where more Jews gathered, chanting anti-Arab slogans. Later that night, a false rumor spread to the predominantly PCI Old City section of Acre that Jamal had been killed.Footnote 2 PCI youths then left the Old City and went to the ethnically mixed city center (HaMerkaz), armed with knives and stones. They vandalized and torched Jewish shops, and Jewish youths responded by torching PCI businesses. The violence persisted for four nights. Approximately 60 people were arrested, including Jamal, the PCI motorist.Footnote 3
My main research design goals were twofold: (1) measure how individuals respond to anger-inducing reminders of intergroup violence in Acre in as controlled a manner as possible; and (2) accurately gauge intergroup attitudes while being sensitive to social desirability concerns. Therefore, I chose to use a behavioral economics experiment. I wanted to see how priming anger about past violence in the residents of Acre would influence their willingness to engage in altruism toward ingroup, outgroup, and ethnically ambiguous partners. My working hypothesis based on the previous literature was that priming anger about past intergroup violence would strengthen ingroup identification and increase discrimination.
I was excited to arrive in Acre, Israel, in May 2011. I had just finished the third year of my PhD studies at New York University and was starting fieldwork for what I hoped would be my job market paper. The previous summer, I had completed fieldwork and a series of behavioral-economics experiments in Southern Israel researching how exposure to violence and anger influenced intragroup altruism in affected Jewish communities.Footnote 4 I planned to extend my research to explore the effects of emotions and exposure to violence on both intergroup and intragroup conflict.
Acre was not the first-choice location for my research. Originally, I wanted to examine ethnic tensions related to Jewish settlements in Palestinian East Jerusalem. However, when I discussed this in a meeting with the director of a respected Palestinian survey company her response was, “Imagine seeing your neighbors’ houses being confiscated or being worried that you would lose your residence permit.Footnote 5 Now trying knocking on that same door to ask political questions.”Footnote 6 The director further explained that she had conducted research in Hebron, Gaza, and other tense areas—but East Jerusalem was always difficult and had extremely low response rates. Thus, the concerns were that it would be difficult to convince Palestinians to participate and that it would be unethical to conduct the research in any meaningful way in East Jerusalem.
Challenging contexts can stymie the most careful of research designs. This is especially true in ethnically tense areas such as Acre. Convincing people to participate in research, difficulties in navigating ethnic and gender issues, and subjects’ responses usually are viewed as challenges to objective, empirical research. Implementation issues typically end up in footnotes or appendices. Yet, the issues that are treated as difficulties or hurdles to field research—hard-to-reach, distrustful populations, and competing narratives around violence—can provide information directly relevant to the research question and cases (Romano Reference Romano2006; Wood Reference Wood2006).Footnote 7 For example, certain populations’ reticence to participate in research may be a direct result of feelings of marginalization that are both a consequence of past violence and a contributing factor to future violence, thereby directly relevant to the research question.
Yet, the issues that are treated as difficulties or hurdles to field research—hard-to-reach, distrustful populations, and competing narratives around violence—can provide information directly relevant to the research question and cases (Romano Reference Romano2006; Wood Reference Wood2006).
WHAT HAPPENED IN ACRE IN 2008?
I needed to know what transpired during the Yom Kippur Riots for two reasons. First, it was necessary to understand the riots as an instance of political violence—how did it start, how did the riots unfold, who participated, and how many people were injured? Second, my research design called for priming the anger about the riots. Therefore, from an experimental control perspective, it was necessary to “get the story right.” Whereas the facts were largely undisputed—(1) Jamal drove into a Jewish neighborhood on Yom Kippur, (2) he was attacked, and (3) then four nights of ethnic rioting ensued—other key details remained contested.
From a series of interviews with local PCIs and Jews, including some who participated in the riots, there were conflicting narratives. Was Jamal drunk and blaring music from his car?Footnote 8 Or was he just driving innocuously? Others questioned whether the initial attack on Jamal had anything to do with Yom Kippur, instead claiming it was a pretext for Jewish youths—angered over a previous drug deal gone bad—to take out their aggression. PCIs and Jews also had drastically different views of the state’s role in the riots. PCIs claimed that the police were biased in favor of Jews,Footnote 9 whereas Jews, particularly those more right-leaning, claimed that the police “did nothing to stop the Arabs.”Footnote 10
These different narratives influence how people viewed the riots. When conducting an experimental prime that aimed to induce respondents’ anger about the riots, these differing interpretations could threaten experimental control.Footnote 11 However, they also were a fundamental part of political violence. For instance other scholars have found disagreement about precipitating incidents that touched off riots in Nigeria (Scacco Reference Scacco2010) and India (Brass Reference Brass1997). My interviews and previous research also suggest that the lines between political and criminal violence are not always clear (Kalyvas Reference Kalyvas2003; Lessing Reference Lessing2015). Different narrative responses to an experimental prime challenge the idea of perfect experimental control in a place like Acre. However, they also point out a possible shortcoming to the conventional wisdom that reminders of ethnic riots create coherent narratives of outgroup aggression and increase discrimination and future ethnic conflict.
ETHNIC IDENTITY AND GENDER
I faced two key issues in the research design and implementation stage that were related to ethnic identity and gender. The names used for the city—Akko (Hebrew), Akka (Arabic), and the more neutral Acre (English)—can be perceived as a political statement. The same is true when referring to the Palestinian residents of Acre as Arab Israelis (used by Jews) or Palestinian Citizens of Israel (PCIs)—the preferred self- identification of many (Rabinowitz and Abu-Baker Reference Rabinowitz and Abu-Baker2005).Footnote 12
My budget permitted me to sample approximately 350 respondents.Footnote 13 I chose three neighborhoods in Acre with varying percentages of Jews and PCIs as my principal sampling frame. The neighborhoods that were selected (figure 1) included the PCI-dominant Old City (97% PCI), the mixed city center HaMerkaz (51.9% Jewish), and the eastern neighborhood of Shuknah Burla (93% Jewish) (Israel Central Bureau of Statistics 2011). These neighborhoods were chosen for two reasons. First, they provided important sample variation across ethnicity. Second, they provided a way to subtly manipulate the ethnic identity of subjects’ group status (i.e., ingroup, outgroup, or unknown) by indicating that their partner with whom they would be splitting money came from “neighborhood X.”Footnote 14
Figure 1 Map of Sample Frame and Enumeration Areas in Acre
I planned to sample proportionally the population across the three neighborhoods: 100 respondents in the Old City, 150 respondents in HaMerkaz (i.e., 75 Jews and 75 PCIs), and 100 respondents in Shuknah Burla. The surveys would take place in respondents’ homes and be conducted in person, with enumerators matching their ethnicity. I used my connections with the local Western Galilee College to recruit a mixed team of PCI and Jewish enumerators.
The ethnically mixed HaMerkaz presented a challenge. Apartment buildings and neighborhoods were intermixed with PCIs and Jews. Given the sensitive nature of the survey and the behavioral-economics experiment—and to prevent any bias—I wanted to ensure that Jewish enumerators surveyed Jewish households and PCI enumerators surveyed PCI households. Language also was an issue. Many PCIs speak both Hebrew and Arabic but the latter was their more comfortable language. We created a workaround to identify Jewish and PCI households using other clues, such as surnames on mailboxes and mezuzahs on Jewish doorposts.
A second issue was gender. Many religious Jewish and Muslim families ascribe to traditional gender roles. After chatting with locals, I was concerned that conservative females would not be willing to be interviewed by a male enumerator, especially because the interview took place in the evening.Footnote 15 Thus, I made a decision to recruit only males. I justified this by the fact that males carry out most acts of violence (Archer Reference Archer2004; Urdal Reference Urdal2006).
THE DIFFICULTY OF SAMPLING
Once I had designed my sample, the key issue that remained was implementing it. Each neighborhood was divided into roughly three equal enumeration areas (EAs). Each enumerator was provided a map of each EA, sampled proportionally to the population, selecting every fourth household within an EA. Respondents were told they could earn between 20 and 76 shekels (NIS; approximately 5 to 20 USD). On average, they earned 52.7 NIS (15.28 USD) for a 20- to 30-minute survey on community relations in Acre.Footnote 16
However, it quickly became clear that there were significant differences in response rates between Jews and PCIs. The Jewish response rate was much lower (i.e., <20% for Jews versus approximately 74% for PCIs). This low response rate among Jews likely had many causes. First, compared to Jews, PCIs exhibited higher levels of trust and denser social ties within their neighborhood.Footnote 17 Second, Jews in Acre had higher levels of income and formal employment (Israel Central Bureau of Statistics 2011), so there were different opportunity costs for participation of Jews versus PCIs. Third, many of the Jews in Acre are immigrants from the former Soviet Union. Given the Soviet legacy of discrimination against Jews, they were likely wary of people knocking on their door asking about their political views (Persky and Berman Reference Persky and Berman2005).Footnote 18
I was able to recruit only 40 Jewish respondents via door-to-door sampling so I decided to employ a professional survey company, the Mahshov Institute, which would contact remaining Jewish households from a landline sampling frame, schedule appointments, and provide translators (mostly from Russian to Hebrew) when needed. The experimental protocol proceeded exactly the same as door-to-door sampling, with subjects being interviewed in their home in person—the only difference was the mode of contact. However this difference complicated the analysis. Would Jewish respondents respond the same via telephone as they would in the face-to-face sample? Furthermore, any differences because of contact mode also might be confounded by the fact that those willing to be interviewed in person would be fundamentally different (i.e., they were more likely to be formally employed).
These differences in sampling and contacting Jews compared to PCIs also shed light on different levels of social capital and trust (i.e., higher among PCIs), as well as socioeconomic status between groups (i.e., higher among Jews) and heterogeneity within the Jewish population in Acre.
These differences in sampling and contacting Jews compared to PCIs also shed light on different levels of social capital and trust (i.e., higher among PCIs), as well as socioeconomic status between groups (i.e., higher among Jews) and heterogeneity within the Jewish population in Acre. All of these factors are important for understanding intergroup conflict in Acre—in particular, the finding of higher levels of ingroup altruism among PCIs compared to Jews.
INTERPRETING THE BEHAVIORAL-ECONOMICS EXPERIMENTS
I used a behavioral-economics experiment to study how reminders of anger about the 2008 Yom Kippur Riots (versus a neutral condition) influenced how much money subjects allocated to three different partners from different neighborhoods meant to signal ethnicity.
For the experiment, participants were told that they would begin with 20 NIS and their partner with 40 NIS. For every 5 NIS they took from their partner, they could keep 2 NIS. To increase the salience of each choice, they were told that they would be randomly paid for one of the choices. Previous research in social psychology and behavioral economics (Brewer Reference Brewer1979; Hoffman, McCabe, and Smith 2006) points to a focal choice of slightly more than a 50–50 split in favor of the participant making the decision. Deviations from this strategy, in which subjects allocate higher amounts to partners from ingroup and lower amounts to those from outgroup, suggest norms of discrimination and high levels of intergroup conflict (Whitt and Wilson Reference Whitt and Wilson2007). Another important part of the game was that the more money “taken” from the partner, the more the total amount of income in the game shrinks, mirroring the “costly” nature of conflict (Bosman, Sutter, and van Winden Reference Bosman, Sutter and van Winden2005).
For the behavioral-economics experiment to work, subjects had to be willing to recognize that (1) neighborhood was a proxy for ethnic identity, and (2) choices in the behavioral-economics experiment reflect attitudes toward ingroup/outgroup members. The results showed that Jews (PCIs) engaged in discrimination, allowing partners in their ingroup neighborhoodsFootnote 19 to keep 12.5 NIS (14.5 NIS) more relative to PCIs (Jews).Footnote 20 However, some respondents had issues with the experimental setup, believing that they were not going to be paid or that their choices were actually going to a respondent. We assured them that they would be paid and that there was no deception.
Another issue was treated as a curious anecdote. A few hardline PCI and Jewish respondents verbally stated and one Jewish respondent wrote on his decision form that they refused to take money from their enemy and gave the full 40 NIS to their partner. The right-wing Jewish respondent said, “I do not want to take any of their dirty Arab drug money.” This decision would be perceived as extremely altruistic to outgroup members but, in fact, the reasons for not taking the money were the opposite. These individuals were only a small minority of the sample and, on average, there was still significant discrimination in Acre. Nevertheless, this anecdote shows how respondents may view constructs differently from the assumed behavioral model. Particularly for hardliners, behavioral experiments may obscure moral reasoning that does not conform to experimental expectations. Another anecdote, expressed by a PCI, suggests a more nuanced response to the riots: “The Jewish guys should have just gone after the driver (Jamal)—he was in the wrong (for driving into their neighborhood on Yom Kippur). But if one (PCI) guy does something wrong, not all Palestinians and Arabs are responsible. The Jewish response was not proportional.”
In behavioral economics experiments these anecdotes would be lost or relegated to footnotes. Yet, they suggest that narratives surrounding riots and how people think about intergroup relations and conflict versus altruism are more nuanced than suggested by the high levels of discrimination in the behavioral-economics experiment.
REACTIONS AND SUGGESTIONS GOING FORWARD
In the fall of 2012, I went on the job market and presented the results from my study in Acre. The main finding was that priming anger about the riots did not increase discrimination. Rather, it reduced the money (i.e., altruism) to all groups (i.e., ingroup and outgroup members). This was contrary to what I previously hypothesized and what the previous literature argues—that is, reminders of past ethnic violence should harden identities and increase conflict (Bar-Tal Reference Bar-Tal2000; Canetti-Nisim et al. Reference Canetti-Nisim, Halperin, Sharvit and Hobfoll2009; Horowitz Reference Horowitz2001). However, my findings echoed the more varied responses to the survey prime and behavioral economics experiment. Attitudes related to the riots and blame attribution are more complicated than conventional wisdom suggests. When presenting my findings at several interviews, people seemed interested in the case and design, but many criticized and correctly pointed out the shortcomings of the sampling, design, and questions about the interpretation of results.
The critiques were fair but I also think that they highlight a shortcoming in how we think about field experiments in difficult places. The obstacles to sampling, ethnic tensions, and interpreting research design usually are seen as barriers to objective estimates of treatment effects and measurement. The process of how we recruit and arrive at these estimates is viewed as a secondary interest. Many implementation issues are relegated to footnotes or barely mentioned in acknowledgments. Yet, these hurdles can inform and contextualize our cases and the study of conflicts.
Scholars should address how research difficulties can provide answers to their research puzzles.
Scholars should address how research difficulties can provide answers to their research puzzles. I offer the following three guidelines for how authors can be more transparent when conducting research in difficult places with histories of conflict:
(1) Be transparent about and document sampling difficulties. Factors that make sampling difficult—low social trust, marginalization, and past exposure to violence—are closely related to many of the questions scholars care about.
(2) Highlight issues with measurement and construct validity. Did participants have issues with standard measurement scales (e.g., social trust, altruism, and social desirability bias)? Were alternative measurement strategies piloted?
(3) Provide other context-specific difficulties. Did local officials try to block the research? Were public and private narratives about the research question similar or different? Answers to these questions should be related to the question of interest.
Many of the places where we want to conduct research are precisely those areas that are difficult to operate in due to conflict or other development issues (Lupu and Michelitch Reference Lupu and Michelitch2018; Weidmann Reference Weidmann2016). Being open about the difficulties in conducting research is an important complement to standard empirical reporting and aligns with the recent movement for improved research transparency in the social sciences more generally (Nosek et al. Reference Nosek, George, Banks, Borsboom, Bowman, Breckler and Buck2015). Finding a way to document, systematize, and be more open about these issues can improve the quality of inferences made as well as provide important information on understanding the causes and consequences of conflict.
